Artificial Intelligence Nanodegree

Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('dogImages/train')
valid_files, valid_targets = load_dataset('dogImages/valid')
test_files, test_targets = load_dataset('dogImages/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
Using TensorFlow backend.
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Here are the different dog breeds in the dataset

In [3]:
dog_names
Out[3]:
['Affenpinscher',
 'Afghan_hound',
 'Airedale_terrier',
 'Akita',
 'Alaskan_malamute',
 'American_eskimo_dog',
 'American_foxhound',
 'American_staffordshire_terrier',
 'American_water_spaniel',
 'Anatolian_shepherd_dog',
 'Australian_cattle_dog',
 'Australian_shepherd',
 'Australian_terrier',
 'Basenji',
 'Basset_hound',
 'Beagle',
 'Bearded_collie',
 'Beauceron',
 'Bedlington_terrier',
 'Belgian_malinois',
 'Belgian_sheepdog',
 'Belgian_tervuren',
 'Bernese_mountain_dog',
 'Bichon_frise',
 'Black_and_tan_coonhound',
 'Black_russian_terrier',
 'Bloodhound',
 'Bluetick_coonhound',
 'Border_collie',
 'Border_terrier',
 'Borzoi',
 'Boston_terrier',
 'Bouvier_des_flandres',
 'Boxer',
 'Boykin_spaniel',
 'Briard',
 'Brittany',
 'Brussels_griffon',
 'Bull_terrier',
 'Bulldog',
 'Bullmastiff',
 'Cairn_terrier',
 'Canaan_dog',
 'Cane_corso',
 'Cardigan_welsh_corgi',
 'Cavalier_king_charles_spaniel',
 'Chesapeake_bay_retriever',
 'Chihuahua',
 'Chinese_crested',
 'Chinese_shar-pei',
 'Chow_chow',
 'Clumber_spaniel',
 'Cocker_spaniel',
 'Collie',
 'Curly-coated_retriever',
 'Dachshund',
 'Dalmatian',
 'Dandie_dinmont_terrier',
 'Doberman_pinscher',
 'Dogue_de_bordeaux',
 'English_cocker_spaniel',
 'English_setter',
 'English_springer_spaniel',
 'English_toy_spaniel',
 'Entlebucher_mountain_dog',
 'Field_spaniel',
 'Finnish_spitz',
 'Flat-coated_retriever',
 'French_bulldog',
 'German_pinscher',
 'German_shepherd_dog',
 'German_shorthaired_pointer',
 'German_wirehaired_pointer',
 'Giant_schnauzer',
 'Glen_of_imaal_terrier',
 'Golden_retriever',
 'Gordon_setter',
 'Great_dane',
 'Great_pyrenees',
 'Greater_swiss_mountain_dog',
 'Greyhound',
 'Havanese',
 'Ibizan_hound',
 'Icelandic_sheepdog',
 'Irish_red_and_white_setter',
 'Irish_setter',
 'Irish_terrier',
 'Irish_water_spaniel',
 'Irish_wolfhound',
 'Italian_greyhound',
 'Japanese_chin',
 'Keeshond',
 'Kerry_blue_terrier',
 'Komondor',
 'Kuvasz',
 'Labrador_retriever',
 'Lakeland_terrier',
 'Leonberger',
 'Lhasa_apso',
 'Lowchen',
 'Maltese',
 'Manchester_terrier',
 'Mastiff',
 'Miniature_schnauzer',
 'Neapolitan_mastiff',
 'Newfoundland',
 'Norfolk_terrier',
 'Norwegian_buhund',
 'Norwegian_elkhound',
 'Norwegian_lundehund',
 'Norwich_terrier',
 'Nova_scotia_duck_tolling_retriever',
 'Old_english_sheepdog',
 'Otterhound',
 'Papillon',
 'Parson_russell_terrier',
 'Pekingese',
 'Pembroke_welsh_corgi',
 'Petit_basset_griffon_vendeen',
 'Pharaoh_hound',
 'Plott',
 'Pointer',
 'Pomeranian',
 'Poodle',
 'Portuguese_water_dog',
 'Saint_bernard',
 'Silky_terrier',
 'Smooth_fox_terrier',
 'Tibetan_mastiff',
 'Welsh_springer_spaniel',
 'Wirehaired_pointing_griffon',
 'Xoloitzcuintli',
 'Yorkshire_terrier']

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Here is the first human image file, for example:

In [4]:
human_files[0]
Out[4]:
'lfw/Gene_Robinson/Gene_Robinson_0004.jpg'

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline   
In [4]:
# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 3

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [5]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer: Of the first 100 images of humans, 99% got a human detected in them. Unfortunately, of the first 100 images of dogs, 11% got a human detected in them, which was not really the case. Some dogs are mistaken as humans.

In [34]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]

## Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.
human_files_with_faces = 0
dog_files_with_faces = 0

for human_file in human_files_short:
        if face_detector(human_file):
            human_files_with_faces += 1
print('Percentage of human images with a detected human face:', human_files_with_faces/len(human_files_short))
            
for dog_file in dog_files_short:
        if face_detector(dog_file):
            dog_files_with_faces += 1     
print('Percentage of dog images with a detected human face:', dog_files_with_faces/len(dog_files_short))            
Percentage of human images with a detected human face: 0.99
Percentage of dog images with a detected human face: 0.11

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer: In my opinion, it is a reasonable expectation to pose on the user that we accept human images only when they provide a clear view of a face. The purpose of our fun app is to tell which dog breed's face looks the most like the human's face, and not which dog breed's silhouette looks the most like the human's silhouette or body part. Haar cascades for face detection are an appropriate technique for human detection in this case. The 11% human detection error rate when it's a dog is kind of high, but still acceptable for our fun app's purpose. We could also use a CNN to detect humans or human faces.

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

In [9]:
## (Optional) TODO: Report the performance of another  
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [6]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [7]:
from keras.preprocessing import image                  
from tqdm import tqdm_notebook as tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [8]:
from keras.applications.resnet50 import preprocess_input as preprocess_input_resnet50, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input_resnet50(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [9]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151)) 

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer: Of the first 100 images of humans, 0% got a dog detected in them. Of the first 100 images of dogs, 100% got a dog detected in them. The Dog Detector looks super accurate.

In [44]:
### Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
human_files_with_dogs = 0
dog_files_with_dogs = 0

for human_file in human_files_short:
        if dog_detector(human_file):
            human_files_with_dogs += 1
print('Percentage of human images with a detected dog:', human_files_with_dogs/len(human_files_short))
            
for dog_file in dog_files_short:
        if dog_detector(dog_file):
            dog_files_with_dogs += 1     
print('Percentage of dog images with a detected dog:', dog_files_with_dogs/len(dog_files_short))            
Percentage of human images with a detected dog: 0.0
Percentage of dog images with a detected dog: 1.0

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [10]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255



(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer: I chose basically the same architecture as in this lesson's example: https://github.com/udacity/aind2-cnn/blob/master/cifar10-classification/cifar10_cnn.ipynb.

I changed the final layer's number of nodes from 10 in the CIFAR-10 lesson to 133 to get a prediction vector that corresponds to the number of possible dog breeds for this project.

Since the tensors in this project are 224 x 224 x 3 rather than the 32 x 32 x 3 tensors from the lesson, it initially gave me around 25M parameters rather than the 500K parameters in the lesson. To reduce the number of parameters, I changed the pool size of the pooling layers from 2 to 3, which then gave me around 2M parameters. To reduce the number of parameters even more, I removed the model.add(Dense(500, activation='relu')) layer and ended up with around 500K parameters.

I also like this CNN architecture because it includes 2 Dropout layers to avoid overfitting.

Create and Configure Augmented Image Generator

In [10]:
from keras.preprocessing.image import ImageDataGenerator

# create and configure augmented image generator
datagen_train = ImageDataGenerator(
    width_shift_range=0.2,  # randomly shift images horizontally (10% of total width)
    height_shift_range=0.2,  # randomly shift images vertically (10% of total height)
    horizontal_flip=True) # randomly flip images horizontally

# create and configure augmented image generator
datagen_valid = ImageDataGenerator(
    width_shift_range=0.2,  # randomly shift images horizontally (10% of total width)
    height_shift_range=0.2,  # randomly shift images vertically (10% of total height)
    horizontal_flip=True) # randomly flip images horizontally

# fit augmented image generator on data
datagen_train.fit(train_tensors)
datagen_valid.fit(valid_tensors)

Visualize Original and Augmented Images

In [22]:
import matplotlib.pyplot as plt

# take subset of training data
x_train_subset = train_tensors[:6]

# visualize subset of training data
fig = plt.figure(figsize=(20,2))
for i in range(0, len(x_train_subset)):
    ax = fig.add_subplot(1, 6, i+1)
    ax.imshow(x_train_subset[i])
fig.suptitle('Subset of Original Training Images', fontsize=20)
plt.show()

# visualize augmented images
fig = plt.figure(figsize=(20,2))
for x_batch in datagen_train.flow(x_train_subset, batch_size=6):
    for i in range(0, 6):
        ax = fig.add_subplot(1, 6, i+1)
        ax.imshow(x_batch[i])
    fig.suptitle('Augmented Images', fontsize=20)
    plt.show()
    break;

Define the Model Architecture

In [11]:
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, GlobalAveragePooling2D
In [31]:
model = None
model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, padding='same', activation='relu', 
                        input_shape=(224, 224, 3)))
model.add(MaxPooling2D(pool_size=3))
model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=3))
model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=3))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dropout(0.4))
model.add(Dense(133, activation='softmax'))

model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_72 (Conv2D)           (None, 224, 224, 16)      208       
_________________________________________________________________
max_pooling2d_17 (MaxPooling (None, 74, 74, 16)        0         
_________________________________________________________________
conv2d_73 (Conv2D)           (None, 74, 74, 32)        2080      
_________________________________________________________________
max_pooling2d_18 (MaxPooling (None, 24, 24, 32)        0         
_________________________________________________________________
conv2d_74 (Conv2D)           (None, 24, 24, 64)        8256      
_________________________________________________________________
max_pooling2d_19 (MaxPooling (None, 8, 8, 64)          0         
_________________________________________________________________
dropout_11 (Dropout)         (None, 8, 8, 64)          0         
_________________________________________________________________
flatten_7 (Flatten)          (None, 4096)              0         
_________________________________________________________________
dropout_12 (Dropout)         (None, 4096)              0         
_________________________________________________________________
dense_11 (Dense)             (None, 133)               544901    
=================================================================
Total params: 555,445
Trainable params: 555,445
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [67]:
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [12]:
from keras.callbacks import ModelCheckpoint 
In [68]:
### Specify the number of epochs that you would like to use to train the model.

batch_size = 32
epochs = 10

### Do NOT modify the code below this line.

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                               verbose=1, save_best_only=True)

model.fit_generator(datagen_train.flow(train_tensors, train_targets, batch_size=batch_size),
                    steps_per_epoch=train_tensors.shape[0] // batch_size,
                    epochs=epochs, verbose=1, callbacks=[checkpointer],
                    validation_data=datagen_valid.flow(valid_tensors, valid_targets, batch_size=batch_size),
                    validation_steps=valid_tensors.shape[0] // batch_size)
Epoch 1/10
207/208 [============================>.] - ETA: 0s - loss: 4.8593 - acc: 0.0113Epoch 00000: val_loss improved from inf to 4.77260, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 61s - loss: 4.8600 - acc: 0.0113 - val_loss: 4.7726 - val_acc: 0.0228
Epoch 2/10
207/208 [============================>.] - ETA: 0s - loss: 4.6845 - acc: 0.0322Epoch 00001: val_loss improved from 4.77260 to 4.64529, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.6841 - acc: 0.0325 - val_loss: 4.6453 - val_acc: 0.0361
Epoch 3/10
207/208 [============================>.] - ETA: 0s - loss: 4.5231 - acc: 0.0495Epoch 00002: val_loss improved from 4.64529 to 4.52245, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 59s - loss: 4.5225 - acc: 0.0496 - val_loss: 4.5225 - val_acc: 0.0461
Epoch 4/10
207/208 [============================>.] - ETA: 0s - loss: 4.3891 - acc: 0.0629Epoch 00003: val_loss improved from 4.52245 to 4.41536, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.3902 - acc: 0.0630 - val_loss: 4.4154 - val_acc: 0.0785
Epoch 5/10
207/208 [============================>.] - ETA: 0s - loss: 4.2894 - acc: 0.0767Epoch 00004: val_loss improved from 4.41536 to 4.36793, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.2887 - acc: 0.0768 - val_loss: 4.3679 - val_acc: 0.0635
Epoch 6/10
207/208 [============================>.] - ETA: 0s - loss: 4.2135 - acc: 0.0858Epoch 00005: val_loss improved from 4.36793 to 4.34806, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.2132 - acc: 0.0858 - val_loss: 4.3481 - val_acc: 0.0648
Epoch 7/10
207/208 [============================>.] - ETA: 0s - loss: 4.1601 - acc: 0.0930Epoch 00006: val_loss improved from 4.34806 to 4.25312, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.1609 - acc: 0.0929 - val_loss: 4.2531 - val_acc: 0.0922
Epoch 8/10
207/208 [============================>.] - ETA: 0s - loss: 4.0941 - acc: 0.1007Epoch 00007: val_loss improved from 4.25312 to 4.17568, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.0956 - acc: 0.1003 - val_loss: 4.1757 - val_acc: 0.0884
Epoch 9/10
207/208 [============================>.] - ETA: 0s - loss: 4.0468 - acc: 0.1034Epoch 00008: val_loss improved from 4.17568 to 4.15982, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 4.0475 - acc: 0.1033 - val_loss: 4.1598 - val_acc: 0.0722
Epoch 10/10
207/208 [============================>.] - ETA: 0s - loss: 3.9893 - acc: 0.1048Epoch 00009: val_loss improved from 4.15982 to 4.15473, saving model to saved_models/weights.best.from_scratch.hdf5
208/208 [==============================] - 58s - loss: 3.9899 - acc: 0.1048 - val_loss: 4.1547 - val_acc: 0.0859
Out[68]:
<keras.callbacks.History at 0x7f96bfb1bb70>

Load the Model with the Best Validation Loss

In [32]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [33]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 10.0478%

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In [13]:
from keras.applications import VGG16, VGG19, ResNet50, InceptionV3, Xception

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer: I chose basically the same architecture as in this lesson's example: https://github.com/udacity/aind2-cnn/blob/master/transfer-learning/transfer_learning.ipynb.

Since the final layer of the convolutional base of complex CNN architectures like VGG16, InceptionV3 and Xception have a high number of filters, it's better to use a GlobalAveragePooling2D layer as the first layer of the classifier instead of flattening its last layer, to avoid having a too high number of parameters.

I changed the final layer's number of nodes to 133 to get a prediction vector that corresponds to the number of possible dog breeds for this project.

I also changed the input shape of the first layer to (7, 7, 2048), which is the size the the tensor outputted by the final layer of the convolutional base of the Xception CNN architecture, and which serves as input to the dog breed classifier.

In [14]:
### Define your architecture.
def build_classifier(train_cnn_architecture):
    classifier = None
    classifier = Sequential()
    classifier.add(GlobalAveragePooling2D(input_shape=train_cnn_architecture.shape[1:]))
    classifier.add(Dense(133, activation='softmax'))
    return classifier

For each architecture (VGG16, VGG19, ResNet50, InceptionV3, Xception), generate the corresponding bottleneck features, and save the best trained model for each architecture in the saved_models/ folder in the repository. Then, test the best model of each architecture against the test set to select the one with the highest accuracy.

In [21]:
cnn_architectures = {'VGG16':VGG16, 'VGG19':VGG19, 'ResNet50':ResNet50, 'InceptionV3':InceptionV3, 'Xception':Xception}
test_accuracies = {}
for cnn_architecture_name, cnn_architecture_model in cnn_architectures.items():
    print('\n\n\n\n\n\n\n\n\n\nTesting for', cnn_architecture_name,'architecture...')
  
    # Obtain Bottleneck Features
    # Import the CNN architecture Model, with the Final Fully-Connected Layers Removed
    conv_base_cnn_architecture = cnn_architecture_model(weights='imagenet', include_top=False)
    # Feed-forward each train, valid and test 3D feature tensor through the convolution base of the VGG-16 model to get the bottleneck features
    print('Obtaining Bottleneck Features...')
    train_cnn_architecture = [conv_base_cnn_architecture.predict(np.expand_dims(train_tensor, axis=0)) for train_tensor in tqdm(train_tensors)]
    valid_cnn_architecture = [conv_base_cnn_architecture.predict(np.expand_dims(valid_tensor, axis=0)) for valid_tensor in tqdm(valid_tensors)]
    test_cnn_architecture = [conv_base_cnn_architecture.predict(np.expand_dims(test_tensor, axis=0)) for test_tensor in tqdm(test_tensors)]
    # Stack the 3D tensor of each sample to build the 4D tensor to input to the neural network# 
    train_cnn_architecture = np.vstack(train_cnn_architecture)
    valid_cnn_architecture = np.vstack(valid_cnn_architecture)
    test_cnn_architecture = np.vstack(test_cnn_architecture)
    
    # Build classifier architecture
    classifier_cnn_architecture = build_classifier(train_cnn_architecture)
    
    # Compile the model
    classifier_cnn_architecture.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
    
    # Train the model
    checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.'+cnn_architecture_name+'.hdf5', 
                               verbose=1, save_best_only=True)
    classifier_cnn_architecture.fit(train_cnn_architecture, train_targets, 
          validation_data=(valid_cnn_architecture, valid_targets),
          epochs=5, batch_size=32, callbacks=[checkpointer], verbose=1)
    
    # Load the Model with the Best Validation Loss
    classifier_cnn_architecture.load_weights('saved_models/weights.best.'+cnn_architecture_name+'.hdf5')
    
    # Test model
    predictions_cnn_architecture = [np.argmax(classifier_cnn_architecture.predict(np.expand_dims(feature, axis=0))) for feature in test_cnn_architecture]
    test_accuracy = np.sum(np.array(predictions_cnn_architecture)==np.argmax(test_targets, axis=1))/len(predictions_cnn_architecture)
    print(cnn_architecture_name,'test accuracy:', round(test_accuracy,4))
    
    test_accuracies[cnn_architecture_name] = test_accuracy

test_accuracies
Testing for VGG16 architecture...
Obtaining Bottleneck Features...


Train on 6680 samples, validate on 835 samples
Epoch 1/5
6496/6680 [============================>.] - ETA: 0s - loss: 4.7932 - acc: 0.0320Epoch 00000: val_loss improved from inf to 4.58949, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 4.7872 - acc: 0.0332 - val_loss: 4.5895 - val_acc: 0.0731
Epoch 2/5
6464/6680 [============================>.] - ETA: 0s - loss: 4.4342 - acc: 0.1055Epoch 00001: val_loss improved from 4.58949 to 4.34466, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 4.4323 - acc: 0.1055 - val_loss: 4.3447 - val_acc: 0.1174
Epoch 3/5
6656/6680 [============================>.] - ETA: 0s - loss: 4.1640 - acc: 0.1629Epoch 00002: val_loss improved from 4.34466 to 4.15551, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 4.1636 - acc: 0.1629 - val_loss: 4.1555 - val_acc: 0.1449
Epoch 4/5
6656/6680 [============================>.] - ETA: 0s - loss: 3.9429 - acc: 0.2169Epoch 00003: val_loss improved from 4.15551 to 3.98725, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 3.9425 - acc: 0.2171 - val_loss: 3.9873 - val_acc: 0.1641
Epoch 5/5
6624/6680 [============================>.] - ETA: 0s - loss: 3.7461 - acc: 0.2446Epoch 00004: val_loss improved from 3.98725 to 3.85119, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 3.7463 - acc: 0.2439 - val_loss: 3.8512 - val_acc: 0.2072
VGG16 test accuracy: 0.2237

Testing for VGG19 architecture...
Obtaining Bottleneck Features...


Train on 6680 samples, validate on 835 samples
Epoch 1/5
6592/6680 [============================>.] - ETA: 0s - loss: 4.8085 - acc: 0.0312Epoch 00000: val_loss improved from inf to 4.65888, saving model to saved_models/weights.best.VGG19.hdf5
6680/6680 [==============================] - 1s - loss: 4.8074 - acc: 0.0316 - val_loss: 4.6589 - val_acc: 0.0491
Epoch 2/5
6464/6680 [============================>.] - ETA: 0s - loss: 4.5187 - acc: 0.0804Epoch 00001: val_loss improved from 4.65888 to 4.44986, saving model to saved_models/weights.best.VGG19.hdf5
6680/6680 [==============================] - 1s - loss: 4.5152 - acc: 0.0814 - val_loss: 4.4499 - val_acc: 0.0994
Epoch 3/5
6560/6680 [============================>.] - ETA: 0s - loss: 4.2910 - acc: 0.1313Epoch 00002: val_loss improved from 4.44986 to 4.27999, saving model to saved_models/weights.best.VGG19.hdf5
6680/6680 [==============================] - 1s - loss: 4.2899 - acc: 0.1319 - val_loss: 4.2800 - val_acc: 0.1353
Epoch 4/5
6592/6680 [============================>.] - ETA: 0s - loss: 4.0968 - acc: 0.1723Epoch 00003: val_loss improved from 4.27999 to 4.12671, saving model to saved_models/weights.best.VGG19.hdf5
6680/6680 [==============================] - 1s - loss: 4.0955 - acc: 0.1722 - val_loss: 4.1267 - val_acc: 0.1569
Epoch 5/5
6592/6680 [============================>.] - ETA: 0s - loss: 3.9270 - acc: 0.1986Epoch 00004: val_loss improved from 4.12671 to 4.00867, saving model to saved_models/weights.best.VGG19.hdf5
6680/6680 [==============================] - 1s - loss: 3.9254 - acc: 0.1991 - val_loss: 4.0087 - val_acc: 0.1689
VGG19 test accuracy: 0.1687

Testing for ResNet50 architecture...
Obtaining Bottleneck Features...


Train on 6680 samples, validate on 835 samples
Epoch 1/5
6656/6680 [============================>.] - ETA: 0s - loss: 4.9784 - acc: 0.0123Epoch 00000: val_loss improved from inf to 4.90333, saving model to saved_models/weights.best.ResNet50.hdf5
6680/6680 [==============================] - 1s - loss: 4.9786 - acc: 0.0123 - val_loss: 4.9033 - val_acc: 0.0120
Epoch 2/5
6304/6680 [===========================>..] - ETA: 0s - loss: 4.8987 - acc: 0.0173Epoch 00001: val_loss improved from 4.90333 to 4.89421, saving model to saved_models/weights.best.ResNet50.hdf5
6680/6680 [==============================] - 0s - loss: 4.8954 - acc: 0.0180 - val_loss: 4.8942 - val_acc: 0.0204
Epoch 3/5
6560/6680 [============================>.] - ETA: 0s - loss: 4.8518 - acc: 0.0203Epoch 00002: val_loss improved from 4.89421 to 4.85317, saving model to saved_models/weights.best.ResNet50.hdf5
6680/6680 [==============================] - 0s - loss: 4.8532 - acc: 0.0201 - val_loss: 4.8532 - val_acc: 0.0144
Epoch 4/5
6624/6680 [============================>.] - ETA: 0s - loss: 4.8172 - acc: 0.0231Epoch 00003: val_loss improved from 4.85317 to 4.81091, saving model to saved_models/weights.best.ResNet50.hdf5
6680/6680 [==============================] - 0s - loss: 4.8182 - acc: 0.0229 - val_loss: 4.8109 - val_acc: 0.0251
Epoch 5/5
6560/6680 [============================>.] - ETA: 0s - loss: 4.7907 - acc: 0.0276Epoch 00004: val_loss improved from 4.81091 to 4.79907, saving model to saved_models/weights.best.ResNet50.hdf5
6680/6680 [==============================] - 0s - loss: 4.7904 - acc: 0.0275 - val_loss: 4.7991 - val_acc: 0.0275
ResNet50 test accuracy: 0.0311

Testing for InceptionV3 architecture...
Obtaining Bottleneck Features...


Train on 6680 samples, validate on 835 samples
Epoch 1/5
6656/6680 [============================>.] - ETA: 0s - loss: 1.2551 - acc: 0.6932Epoch 00000: val_loss improved from inf to 0.61499, saving model to saved_models/weights.best.InceptionV3.hdf5
6680/6680 [==============================] - 3s - loss: 1.2528 - acc: 0.6936 - val_loss: 0.6150 - val_acc: 0.8072
Epoch 2/5
6624/6680 [============================>.] - ETA: 0s - loss: 0.4855 - acc: 0.8507Epoch 00001: val_loss improved from 0.61499 to 0.60875, saving model to saved_models/weights.best.InceptionV3.hdf5
6680/6680 [==============================] - 2s - loss: 0.4849 - acc: 0.8506 - val_loss: 0.6087 - val_acc: 0.8335
Epoch 3/5
6624/6680 [============================>.] - ETA: 0s - loss: 0.3529 - acc: 0.8909Epoch 00002: val_loss did not improve
6680/6680 [==============================] - 2s - loss: 0.3545 - acc: 0.8900 - val_loss: 0.6102 - val_acc: 0.8431
Epoch 4/5
6656/6680 [============================>.] - ETA: 0s - loss: 0.2753 - acc: 0.9096Epoch 00003: val_loss did not improve
6680/6680 [==============================] - 2s - loss: 0.2751 - acc: 0.9096 - val_loss: 0.6236 - val_acc: 0.8491
Epoch 5/5
6496/6680 [============================>.] - ETA: 0s - loss: 0.2157 - acc: 0.9290Epoch 00004: val_loss did not improve
6680/6680 [==============================] - 2s - loss: 0.2144 - acc: 0.9296 - val_loss: 0.6322 - val_acc: 0.8467
InceptionV3 test accuracy: 0.7931

Testing for Xception architecture...
Obtaining Bottleneck Features...


Train on 6680 samples, validate on 835 samples
Epoch 1/5
6656/6680 [============================>.] - ETA: 0s - loss: 1.1759 - acc: 0.7225Epoch 00000: val_loss improved from inf to 0.52864, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 6s - loss: 1.1729 - acc: 0.7229 - val_loss: 0.5286 - val_acc: 0.8395
Epoch 2/5
6656/6680 [============================>.] - ETA: 0s - loss: 0.4163 - acc: 0.8727Epoch 00001: val_loss improved from 0.52864 to 0.48623, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 4s - loss: 0.4165 - acc: 0.8729 - val_loss: 0.4862 - val_acc: 0.8515
Epoch 3/5
6656/6680 [============================>.] - ETA: 0s - loss: 0.3198 - acc: 0.8992Epoch 00002: val_loss improved from 0.48623 to 0.47644, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 4s - loss: 0.3197 - acc: 0.8991 - val_loss: 0.4764 - val_acc: 0.8551
Epoch 4/5
6656/6680 [============================>.] - ETA: 0s - loss: 0.2634 - acc: 0.9151Epoch 00003: val_loss did not improve
6680/6680 [==============================] - 4s - loss: 0.2632 - acc: 0.9151 - val_loss: 0.4802 - val_acc: 0.8563
Epoch 5/5
6656/6680 [============================>.] - ETA: 0s - loss: 0.2216 - acc: 0.9310Epoch 00004: val_loss did not improve
6680/6680 [==============================] - 4s - loss: 0.2219 - acc: 0.9310 - val_loss: 0.4869 - val_acc: 0.8491
Xception test accuracy: 0.8517
Out[21]:
{'InceptionV3': 0.7930622009569378,
 'ResNet50': 0.031100478468899521,
 'VGG16': 0.22368421052631579,
 'VGG19': 0.1686602870813397,
 'Xception': 0.85167464114832536}

After training on 5 epochs each CNN architecture, Xception has the highest accuracy on the test set, around 85%

In [15]:
classifier_Xception = Sequential()
classifier_Xception.add(GlobalAveragePooling2D(input_shape= (7, 7, 2048)))
classifier_Xception.add(Dense(133, activation='softmax'))

classifier_Xception.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_1 ( (None, 2048)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 133)               272517    
=================================================================
Total params: 272,517
Trainable params: 272,517
Non-trainable params: 0
_________________________________________________________________
In [16]:
# Load the Model with the Best Test Accuracy
classifier_Xception.load_weights('saved_models/weights.best.Xception.hdf5')

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

In [17]:
### Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.
from keras.applications.xception import Xception, preprocess_input as preprocess_input_xception

def Xception_predict_breed(img_path):
    # extract bottleneck features from original image by passing through Xception convolutional base
    tensor = path_to_tensor(img_path)
    bottleneck_feature = Xception(weights='imagenet', include_top=False).predict(preprocess_input_xception(tensor))
    # obtain predicted vector by passing bottleneck features through classifier
    predicted_vector = classifier_Xception.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Test predicting the dog breed on an image

In [18]:
Xception_predict_breed('dogImages/test/096.Labrador_retriever/Labrador_retriever_06472.jpg')
Out[18]:
'Labrador_retriever'

Test Dog Breed Classifier on dog images

In [64]:
x_test = test_files[:30]
y_test = test_targets[:30]

y_hat = []
for img_path in x_test:
    prediction = Xception_predict_breed(img_path)
    y_hat.append(prediction)
In [65]:
# plot test images, their predicted labels, and ground truth
fig = plt.figure(figsize=(20, 20))
for i, img_path in enumerate(x_test):
    ax = fig.add_subplot(6, 5, i + 1, xticks=[], yticks=[])
    x_img = cv2.imread(img_path)
    cv_rgb = cv2.cvtColor(x_img, cv2.COLOR_BGR2RGB)
    plt.imshow(np.squeeze(cv_rgb))
    prediction = y_hat[i]
    true_idx = np.argmax(y_test[i])
    ax.set_title("Predicted: {} \nTrue: {}".format(prediction, dog_names[true_idx]),
                 color=("green" if prediction == dog_names[true_idx] else "red"))

Test Dog Breed Classifier on human images

In [55]:
x_test_humans = human_files[:10]

y_hat_humans = []
for img_path in x_test_humans:
    prediction = Xception_predict_breed(img_path)
    y_hat_humans.append(prediction)
In [56]:
# plot test images, their predicted labels
fig = plt.figure(figsize=(20, 20))
for i, img_path in enumerate(x_test_humans):
    ax = fig.add_subplot(6, 5, i + 1, xticks=[], yticks=[])
    x_img = cv2.imread(img_path)
    cv_rgb = cv2.cvtColor(x_img, cv2.COLOR_BGR2RGB)
    plt.imshow(np.squeeze(cv_rgb))
    prediction = y_hat_humans[i]
    ax.set_title(prediction)

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [23]:
### Write your algorithm.
def dog_breed_identifier(img_path):
    isHuman = False
    isDog = False
    
    # Detect human or dog
    if dog_detector(img_path):
        isDog = True
        print('\n\n\n\n\nDog detected!')
    elif face_detector(img_path):
        isHuman = True
        print('\n\n\n\n\nHuman detected!')
    else:
        print('\n\n\n\n\nNo human or dog detected.')
        
    # Show image
    img = cv2.imread(img_path)
    cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    plt.imshow(cv_rgb)
    plt.show()
    
    # Identify dog breed
    predicted_breed = Xception_predict_breed(img_path)
    predicted_breed_split = predicted_breed.split('_')
    predicted_breed_formatted = ""
    for word in predicted_breed_split:
        predicted_breed_formatted += word + " "
    if isDog:
        print('Looks like a', predicted_breed_formatted)
    elif isHuman:
        print('You look like a', predicted_breed_formatted)

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer: I tested the algorithm on 5 dog images and 5 human images.

For the dog images, it accurately detected a dog in all of them. I googled the predicted dog breed for each image and the Dog Breed Predictor seems to have guessed the dog breed accurately in all of them.

For the human images, it also accurately detected a human in all of them. I googled the predicted dog breed for each image and the Dog Breed Predictor seems to have guessed a similar-looking dog breed accurately in all of them.

Here are possible points of improvement for our algorithm:

  • Our human detector detects human faces in around 10% of dog images, which is a high error rate. We could use a binary CNN classifier trained on human-face and non-human-face images to improve its accuracy.
  • Using a CNN to detect human faces, we could also detect human faces viewed from the side if trained with such images
  • Augment the training set to improve performance
  • Make the algorithm work in real time from a video feed rather than from an image file
  • Use a CNN architecture with less parameters than Xception to increase speed, but train that model on more epochs to get similar test accuracy

Test algorithm on dog images

In [24]:
x_test_algo = test_files[-5:]
for img_path in x_test_algo:
    dog_breed_identifier(img_path)




Dog detected!
Looks like a Pekingese 





Dog detected!
Looks like a Bloodhound 





Dog detected!
Looks like a Japanese chin 





Dog detected!
Looks like a Greyhound 





Dog detected!
Looks like a German pinscher 

Test algorithm on human images

In [25]:
x_test_humans_algo = human_files[-5:]
for img_path in x_test_humans_algo:
    dog_breed_identifier(img_path)




Human detected!
You look like a Dachshund 





Human detected!
You look like a Cavalier king charles spaniel 





Human detected!
You look like a Glen of imaal terrier 





Human detected!
You look like a Dachshund 





Human detected!
You look like a Smooth fox terrier 
In [ ]: